def main():
parameters_2_optimize = (list(model_CNN.parameters()) + list(model_my_fc6.parameters()) +
list(model_score_regressor.parameters()))
parameters_2_optimize_named = (list(model_CNN.named_parameters()) + list(model_my_fc6.named_parameters()) +
list(model_score_regressor.named_parameters()))
if with_dive_classification:
parameters_2_optimize = parameters_2_optimize + list(model_dive_classifier.parameters())
parameters_2_optimize_named = parameters_2_optimize_named + list(model_dive_classifier.named_parameters())
if with_caption:
parameters_2_optimize = parameters_2_optimize + list(model_caption.parameters())
parameters_2_optimize_named = parameters_2_optimize_named + list(model_caption.named_parameters())
optimizer = optim.Adam(parameters_2_optimize, lr=0.0001)
# print('Parameters that will be learnt: ', parameters_2_optimize_named)
criterions = {}
criterion_final_score = nn.MSELoss()
penalty_final_score = nn.L1Loss()
criterions['criterion_final_score'] = criterion_final_score
criterions['penalty_final_score'] = penalty_final_score
if with_dive_classification:
criterion_dive_classifier = nn.CrossEntropyLoss()
criterions['criterion_dive_classifier'] = criterion_dive_classifier
if with_caption:
criterion_caption = utils_1.LanguageModelCriterion()
criterions['criterion_caption'] = criterion_caption
train_dataset = VideoDataset('train')
test_dataset = VideoDataset('test')
train_dataloader = DataLoader(train_dataset, batch_size=train_batch_size, shuffle=True)
test_dataloader = DataLoader(test_dataset, batch_size=test_batch_size, shuffle=False)
print('Length of train loader: ', len(train_dataloader))
print('Length of test loader: ', len(test_dataloader))
print('Training set size: ', len(train_dataloader)*train_batch_size,
'; Test set size: ', len(test_dataloader)*test_batch_size)
# actual training, testing loops
for epoch in range(100):
# saving_dir = 'c3davg_140_saved' # ADDED PATH FOR SAVING DIRECTORY
saving_dir = 'c3davg_lstm_encode' # ADDED PATH FOR SAVING DIRECTORY
print('-------------------------------------------------------------------------------------------------------')
for param_group in optimizer.param_groups:
print('Current learning rate: ', param_group['lr'])
train_phase(train_dataloader, optimizer, criterions, epoch)
test_phase(test_dataloader)
if (epoch+1) % model_ckpt_interval == 0: # save models every 5 epochs
save_model(model_CNN, 'model_CNN', epoch, saving_dir)
save_model(model_my_fc6, 'model_my_fc6', epoch, saving_dir)
save_model(model_score_regressor, 'model_score_regressor', epoch, saving_dir)
if with_dive_classification:
save_model(model_dive_classifier, 'model_dive_classifier', epoch, saving_dir)
if with_caption:
save_model(model_caption, 'model_caption', epoch, saving_dir)
def main():
parameters_2_optimize = (list(model_CNN.parameters()) + list(
model_my_fc6.parameters())) + list(model_caption.parameters())
parameters_2_optimize_named = (list(model_CNN.named_parameters()) + list(
model_my_fc6.named_parameters())) + list(
model_caption.named_parameters())
optimizer = optim.Adam(parameters_2_optimize, lr=0.0001)
print('Parameters that will be learnt:', len(parameters_2_optimize_named))
criterions = {}
criterion_hockey_classifier = nn.CrossEntropyLoss()
# criterion_hockey_classifier = nn.MSELoss()
criterions['criterion_hockey_classifier'] = criterion_hockey_classifier
criterion_caption = utils_1.LanguageModelCriterion()
criterions['criterion_caption'] = criterion_caption
train_dataset = VideoDataset('train')
test_dataset = VideoDataset('test')
train_dataloader = DataLoader(train_dataset,
batch_size=train_batch_size,
shuffle=True)
test_dataloader = DataLoader(test_dataset,
batch_size=test_batch_size,
shuffle=False)
print('Length of train loader: ', len(train_dataloader))
print('Length of test loader: ', len(test_dataloader))
print('Training set size: ',
len(train_dataloader) * train_batch_size, '; Test set size: ',
len(test_dataloader) * test_batch_size)
losses = []
# actual training, testing loops
for epoch in range(100):
print(
'---------------------------------------------------------------------------------'
)
for param_group in optimizer.param_groups:
print('Current learning rate: ', param_group['lr'])
loss = train_phase(train_dataloader, optimizer, criterions, epoch)
losses.append(loss.cpu().item())
test_phase(test_dataloader)
if (epoch +
1) % model_ckpt_interval == 0: # save models every 5 epochs
save_model(model_CNN, 'model_CNN', epoch, saving_dir)
save_model(model_my_fc6, 'model_my_fc6', epoch, saving_dir)
save_model(model_caption, 'model_caption', epoch, saving_dir)
plt.plot(range(len(losses)), losses)
plt.savefig('losses5.png')